CameraMaster: Unified Camera Semantic-Parameter Control for Photography Retouching

TL;DR

CameraMaster tackles the difficulty of controllable, physically consistent image retouching by decoupling camera directives from numeric parameters into semantic and parameter streams that are jointly conditioned within a diffusion-transformer backbone. It introduces parameter embeddings and Camera-FiLM-based modulation with cross-attention to achieve monotonic, near-linear responses across multiple camera factors, validated on a 78K-image dataset (CameraSet) with strong quantitative and qualitative results. Ablation studies confirm the importance of both streams and their fusion for precise, multi-parameter control and content fidelity. The work advances practical camera-centric editing by offering scalable, interpretable, and robust cross-parameter composition for diffusion-based retouching.

Abstract

Text-guided diffusion models have greatly advanced image editing and generation. However, achieving physically consistent image retouching with precise parameter control (e.g., exposure, white balance, zoom) remains challenging. Existing methods either rely solely on ambiguous and entangled text prompts, which hinders precise camera control, or train separate heads/weights for parameter adjustment, which compromises scalability, multi-parameter composition, and sensitivity to subtle variations. To address these limitations, we propose CameraMaster, a unified camera-aware framework for image retouching. The key idea is to explicitly decouple the camera directive and then coherently integrate two critical information streams: a directive representation that captures the photographer's intent, and a parameter embedding that encodes precise camera settings. CameraMaster first uses the camera parameter embedding to modulate both the camera directive and the content semantics. The modulated directive is then injected into the content features via cross-attention, yielding a strongly camera-sensitive semantic context. In addition, the directive and camera embeddings are injected as conditioning and gating signals into the time embedding, enabling unified, layer-wise modulation throughout the denoising process and enforcing tight semantic-parameter alignment. To train and evaluate CameraMaster, we construct a large-scale dataset of 78K image-prompt pairs annotated with camera parameters. Extensive experiments show that CameraMaster produces monotonic and near-linear responses to parameter variations, supports seamless multi-parameter composition, and significantly outperforms existing methods.

Figures (11)

• Figure 1: CameraMaster: Unified, monotonic, and near-linear photography control. Our framework enables precise, physically consistent retouching across seven key camera dimensions, setting a new benchmark for controllable image retouching.
• Figure 2: Pipeline of CameraMaster for image retouching. Given an input image, a camera directive $d$ and a content description $c$, CameraMaster decouples control into two streams: a semantic stream that encodes $d$ and $c$, and a parameter stream that parses the directive into a calibrated camera vector $s$ and its embedding $\mathbf{z}_{\text{cam}}$. The parameter embedding $\mathbf{z}_{\text{cam}}$ modulates both the semantic representations ($\mathbf{C}_{\mathrm{raw}}$, $\mathbf{D}_{\mathrm{raw}}$) and the time embedding $\mathbf{t}$, providing explicit parameter awareness. The modulated features are fused into a unified semantic context and injected into the DiT backbone, enabling continuous, monotonic, and physically consistent camera control for image retouching.
• Figure 3: Visual comparison of Color Temperature (CCT) task retouching results across different methods. This method achieves continuous, monotonous, and precise control over retouching intensity, which strikes a balance between following directions and preserving the original image's consistency to meet user needs.
• Figure 4: Examples showcase the cross-parameter retouching power of CameraMaster on the real-world dataset. Our method allows for continuous and fine-grained retouching across multiple parameter combinations, illustrating CameraMaster's universal capability in achieving diverse retouching objectives.
• Figure 5: Visual comparison of Saturation task retouching results across different methods.